Agent Skills Are Having a Moment

A few years ago, organizations were busy collecting prompts. Then they started collecting tools and connectors. Now, attention is shifting toward skills: reusable instruction packages, documentation, domain knowledge, and supporting code that teach an AI agent how to perform a particular kind of work.

That may sound like little more than a turn of the AI terminology wheel. And, there is some of that. Prompt engineering became context engineering, and now much of that context is being organized into agent skills.

But something more substantial is happening beneath the language. AI models already know how to write code, call APIs, and work through complicated technical problems. What they often lack is the knowledge that experienced humans carry around in their heads:

• How your organization’s systems fit together
• Which data sources can be trusted
• What a particular metric means in practice
• Which steps matter in a recurring workflow
• What usually goes wrong
• How the work should be evaluated

A good skill captures some of that expertise and gives it to an agent in a form it can repeatedly access. That makes skills more than saved prompts. At their best, they become small, portable pieces of institutional intelligence.

Recently, we’ve been exploring that idea from a number of different angles. Here’s a roundup.

Why Context Is Becoming the Competitive Advantage

What happens when AI models become good enough to write most of the code themselves? According to Stephen Price, the next competitive advantage won’t come from larger models or longer prompt chains. It’ll come from giving AI agents better context.

In The Rise of the Agentic Analyst, Stephen argues that coding agents are rapidly changing the role of the modern analyst. Today’s agents already know how to write code, use APIs, and work through complex technical problems. What they lack is an understanding of how a particular organization actually works: its business rules, trusted data sources, workflows, and institutional knowledge.

That’s where skills come in. Rather than treating AI as a chatbot with an ever-growing collection of tools, Stephen describes a shift toward reusable skills that package organizational knowledge into forms AI agents can repeatedly apply. Those skills become the bridge between powerful general-purpose models and the specialized expertise that makes enterprise analytics valuable.

The article also explores how recent advances in coding agents are changing the underlying tech stack. As agents become increasingly capable of generating and executing their own code, organizations will need fewer narrowly defined tools and more well-designed context. That shift has implications for the platforms we choose, the way analytics teams work, and the skills analysts themselves will need to develop.

The result is what Stephen calls the Agentic Analyst, a new generation of analytics professionals who combine deep business knowledge with the ability to guide AI agents through increasingly sophisticated analytical work.

Why Skills Are the New Context Layer for Agentic Analytics

In Why Skills Are the New Context Layer for Agentic Analytics, Stephen Price begins with a deceptively simple argument: The next major improvement in AI performance may not come from a larger model or a more complicated tech stack, but from giving agents better context.

As coding agents become increasingly capable, they need fewer narrowly prescribed tools. They can often write the code they need themselves. What they cannot infer is how an organization’s internal systems, data, and business rules relate to one another. Skills help close that gap.

Stephen makes the case that analysts are particularly well-positioned to build such skills. Analysts already understand the tables, definitions, exceptions, and relationships that determine whether an answer is useful or merely plausible.

Writing that knowledge down in a clear, structured, and executable form may be one of the most valuable things an analyst can now do.

Introducing the Query Tableau Data Agent Skill

Stephen’s next article, Introducing the Query Tableau Data Agent Skill, offers a tangible example of what he’s been talking about. Action’s new open-source Query Tableau Data Skill” enables coding agents to explore, understand, and query Tableau environments through the VizQL Data Service API and Tableau’s broader metadata ecosystem.

The skill can help an agent:

• Explore Tableau’s Data Catalog
• Discover workbooks, dashboards, and views
• Understand data lineage and structure
• Query published views and data sources
• Retrieve governed data for further analysis

A Tableau environment contains years of accumulated decisions about measures, calculations, filters, business definitions, and visual relationships. A well-constructed agent skill can make that institutional knowledge more accessible to an AI agent while preserving the context and governance that make the data trustworthy.

The Query Tableau Data Skill is the first release in what will become a larger Action Agent Skills series.

Not All AI Skills Are Equal

Once organizations begin building skills, a predictable problem appears: They start building lots of them. In the latest episode of The Sensemakers, Keith Helfrich, Jonathan Drummey, and Stephen Price examine the difference between a genuinely useful AI skill and a folder full of lightly-edited documentation.

The conversation covers:

• Why domain expertise matters more than AI-generated boilerplate
• How meaning compaction makes skills more useful
• Why a skill should support a focused workflow rather than reproduce an entire API surface
• How governance and marketplaces can help organizations manage proliferating AI skill libraries
• Why quality will matter far more than quantity

The conclusion is simple: A large collection of mediocre skills simply creates a new layer of organizational clutter. The better goal is to capture institutional expertise carefully, compress it thoughtfully, and organize these skills so agents can use them reliably.

Where Enterprise AI Is Headed

If the past few years were about discovering what large language models can do, the next few may be about teaching them what your organization actually knows.

Every business accumulates knowledge that never makes it into a database or work documentation. Analysts know which metrics executives actually trust. Engineers know why one workflow succeeds while another repeatedly fails. Subject-matter experts understand the exceptions, shortcuts, and edge cases that make work possible in the real world.

AI models don’t automatically inherit that knowledge. Skills provide one practical way to capture it, organize it, and make it available to AI agents in a reusable form. They transform institutional expertise from something that exists only in people’s heads into an asset that can be shared across teams and applied consistently.

That’s why we believe skills are becoming one of the most important building blocks in enterprise AI. They’re not replacing people. They’re making human expertise more durable, portable, and ultimately more valuable.

Additional Resources

• Action’s Query Tableau Data Skill
• The Real Meaning of Headless BI
• Headless BI Isn’t About Dashboards. It’s About Architecture
• Are We At Peak MCP?